When a patient begins taking any medication for a length of time, a titration of the amount of drug taken by the patient is necessary in order to achieve the optimal benefit of the drug and at the same time to prevent any undesirable side effects that taking too much of the drug could produce. Thus, there is a continuous balance between taking enough drug in order to gain the benefits from that drug and at the same time not taking so much drug as to illicit a toxic event.
There is large inter-individual variability in the patient pharmacokinetics of drugs. What may be an appropriate drug dose for one individual, may be too much or too little for another. Prior to this invention a physician was required to estimate the correct drug dosage for a patient and then to experiment with that dosage, usually by trial and error, until the correct dosage was achieved. Likewise, the FDA labeling of a drug suggests dosages based on epidemiological studies and again does not account for inter-individual variability. Non-linear least squares modeling methods involve the use of large amounts of data relating to a general population in order to calculate a best fit. Much like linear regression models, this method cannot take into account the variability between people with the same population characteristics.
Bayesian analysis is another method used to relate drug dose to efficacy. This method employs large-scale population parameters to stratify a population in order to better characterize the individuals. This method does not take into account the changes that can occur within a person over time, and as a result cannot reliably estimate dosages.
Pharmacokinetic compartment modeling has had success with some drugs, but because the models are static and cannot adapt themselves to changes within a population or a patient, they are once again undesirable for dynamically determining drug dosages.
Expert systems have been developed using similar technology to predict drug dosages for immunosuppressant drugs (see, e.g., U.S. Pat. Nos. 5,365,948, 5,542,436 and 5,694,950). These algorithms, however, are not generic and only use immunosuppressant blood levels. Each algorithm is specific to an individual immunosuppressant drug. As it stands, these inventions cannot be applied to other drugs and do not have a non-linear feedback loop mechanism.